I have a set of ant tasks that I use to run my test suite, occasionally one of those tests will freeze and my entire test suite will hang. I added a shutdown handler so when I hit Ctrl+C ant will shutdown gracefully and give me a report with the final test marked as not run. (This is important because these are integration tests and can run for hours) This works great except for on Windows where my shutdown hook is not being called. Is there a way I can get ant to respond to any kind of input and do a graceful shutdown?
It seems this is a long-running known issue.
The problem is that on Windows the Ant Ctrl+C is, as you have observed, not propagated to the child VMs. Things you might consider:
Break up the test into smaller pieces and use a timeout to kill anything that hangs. This will limit the data lost to the one test that hangs.
In your test run, add a 'listener' thread that waits for a shutdown 'signal' (perhaps the presence of a flag file) and arrange for that signal to be set by Ant, on command from the console, if hang is detected.
This appears complex, but might be worth a shot. You'd need to combine the Ant parallel and input tasks to run the tests in one thread, and wait for input from the console in a second thread. When abort is chosen, the signal file is written, this is detected in the test run 'listener', causing it to terminate. Any other input would lead to clean termination of the run. The problem with this is that, if the test completes successfully, you're left with Ant waiting for user input, but you could put an overall timeout in place for that. (I've not given an example of how the test run code might detect the signal file.)
Psuedo-Ant:
<property name="signal.abort" value="stop.txt" />
<target name="runner">
<delete file="${signal.abort}" />
<parallel timeout="86400000">
<sequential>
<!-- run tests here -->
</sequential>
<sequential>
<input validargs="y,n"
message="Abort the test (y/n)?"
addproperty="abort.test" />
<condition property="do.abort">
<equals arg1="y" arg2="${abort.test}"/>
</condition>
<ant target="terminator" />
</sequential>
</parallel>
</target>
<target name="terminator" if="do.abort">
<echo message="abort" file="${signal.abort}" />
</target>
Related
I have tasks running in Ant, which I'm quite new to, as part of a CI chain of build events. I used a tutorial to create the file for testing/linting/etc a PHP application.
The first important directives in the build.xml are:
<property name="phpmd" value="phpmd"/>
<property name="phpunit" value="phpunit"/>
This works fine as is, assuming that phpmd/phpunit are on the path, and using phpunit as a further example, is run under the following target:
<target name="phpunit" unless="phpunit.done" depends="prepare" description="Run unit tests with PHPUnit">
<exec executable="${phpunit}" resultproperty="result.phpunit" taskname="phpunit">
<arg value="--configuration"/>
<arg path="${basedir}/phpunit.xml"/>
</exec>
<property name="phpunit.done" value="true"/>
</target>
All this works well as is- but I want to use docker from now on, which I had hoped would simply mean changing <property name="phpunit" value="phpunit"/> to <property name="phpunit" value="docker-compose run php phpunit"/>, but this instead gives me the following error:
Execute failed: java.io.IOException: Cannot run program "docker-compose run -w /var/www/src php phpunit" (in directory "/var/lib/jenkins/jobs/Blah blah blah/workspace/src"): error=2, No such file or directory
I know that you would usually add additional <arg/> nodes to targets- but is it not possible at all to provide the full command with inline arguments on the initial <property>?
Ant is obviously complaining because, along with those inline arguments, that executable doesn't exist. Will I have to use arg nodes and update every single target?
Using docker-compose alone works fine, but I need the args for the correct container and working directory to be used- preferably inline, otherwise I have to insert many arg nodes.
In the end I just created a property for phpunit arguments and then added it by using the <args line="${phpunitArgs}">.
Definitely not ideal, but at least it does what it is supposed to. I certainly prefer using Gulp! XML feels like a bad choice for a build system.
In one of our build script, we have following simple copy task added ->
<copy todir="${targetdir}"
file="${sourcedir}/modules/glassfish.jaxb.xjc_1.0.0.0_2-1-12.jar"/>
This copy task started hanging when the glassfish jar name got changed (version upgrade which are not in our control) at the source location. I was expecting it to error out causing the build failure in that case. Actually at first I was not able to figure out at what particular step build was hanging. Then when I added "-debug" to the ant command and I realized its successfully completing a step prior to this copy task and still there was no trace of copy command that is hung. When I updated the new jar name, it worked fine and build was successful which proved that the copy task is hanging because of filename got changed. To make it easy to debug next time, I added an echo statement like below just prior to that copy task ->
<echo message="Copying glassfish jar to ${targetdir}.."/>
But I am still confused as to why it didn't give error with build failure? I am using Apache Ant version 1.7.1. Could this be a bug? How else I can avoid this situation in future with just the copy task (without using * in the jar name)? TIA
That worked for me. Well, didn't work for me. I got the error message. I am using Ant 1.8 and Ant 1.9.2. I didn't try it with Ant 1.7, but I doubt it's a bug.
Try to use the -v parameter in Ant:
$ ant -v target
And be prepared for a longwinded output. This will give you information what's going on with Ant, and may explain why it's freezing. There's a few things you could do: Use a fileset to specify the file.
<copy todir="${targetdir}">
<fileset dir="${sourcedir}/modules">
<include name="glassfish*.jar"/> <!-- Will catch any glassfish.jar -->
</fileset>
</copy>
Of course, if the file doesn't exist, you won't get an error or even a warning. However, a <fail/> before will detect the issue:
<fail message="missing file ${sourcedir}/modules/glassfish.jaxb.xjc_1.0.0.0_2-1-12.jar">
<condition>
<not>
<available
file="${sourcedir}/modules/glassfish.jaxb.xjc_1.0.0.0_2-1-12.jar"
type="file"/>
</not>
</condition>
</fail>
To force the build to quit, an alternative way
<available file="${sourcedir}/modules/glassfish.jaxb.xjc_1.0.0.0_2-1-12.jar"
property="glassfish.jaxb.xjc.jar.present"/>
<fail message="you message" unless="glassfish.jaxb.xjc.jar.present"/>
just a few lines less :)
If you want to dig into it, try this:
write a simple build file, which contains only one target with copy, and put it to the same place of your main build file.
<target name="test-copy">
<!-- here use an old (wrong) file name -->
<copy todir="${targetdir}"
file="${sourcedir}/modules/glassfish.jaxb.xjc_1.0.0.0_2-1-12.jar"/>
</target>
run it, check if it fails or hangs.
If this simple build file works, it's very possible that something else in your main build file is causing the bug.
I have a STAF job which kicks off an Ant script, all run under a functional ID. The ant script does basic 'init' tasks which work fine, and then hangs on one of the following tasks: delete, mkdir, or junit, depending on how much muddling I do in the file system.
<target name="run.nightly.tests" depends="init">
<delete dir="/path/results/latestDate" />
<mkdir dir="/path/results/latestDate" />
<chmod perm="777" dir="/path/results/latestDate" />
<junit printsummary="on" fork="no">
<!-- typical classpath/tests setup snipped -->
</junit>
</target>
I am able to sudo to this functional ID and run the corresponding commands from a shell just fine (ex: rm -rf /path/results/latestDate). Clearly something is different about running in the Ant environment, under the functional id. (In fact, I can run the script just fine with my own id)
/path/results has 777 permissions before my script runs, and when /path/results/latestDate exists it is owned by the same functional id w/ 777 permissions.
STAF launches this script:
export ANT_HOME=/opt/apache-ant-1.8.2
#This entire directory tree and jar files are world r+x
LIB_DIR=/home/afreed/automation/dependencies/mail
ant -debug -verbose -buildfile nightlyTest.xml -lib ${LIB_DIR}/mail.jar:${LIB_DIR}/activation.jar
I would understand if Ant would fail on any of these tasks with a permissions error but I cannot understand why it would hang.
I would like help either a) determining why there is a hang or b) how to convert the hang to a hard failure
Solved, the Ant script was hanging on an exec command that was waiting for input from STDIN. (Execution had continued past that point because it was launched with 'spawn' attribute.) When I removed this 'exec' task from my 'init' target, the Ant script ran as expected.
<!-- hanging task in init-->
<exec executable="blah.sh" spawn="true">
I want to make a Jenkins job to run an ant task to run all tests in my codebase which are tagged #Ignore because using annotations like #Category(IgnoredTest.class) do not work with our test run parallelization. After a lot of searching it looks undoable, but I still have hope. Help?
JUnit 4.10
I'm not sure what the impediment is with your "test run parallelization", but you might be able to do this with a rule if you're willing to use a custom "ignore" annotation instead of the JUnit one. The reason for that is that JUnit handles #Ignored tests at the Runner level, specifically in the BlockJUnit4ClassRunner.runChild() (by default). If you could find a way to use a custom Runner in Ant, you could come up with one to meet your needs pretty easily, but I don't know if that's easily doable in Ant.
As I first mentioned, though, you can easily use a different annotation and a rule to choose which methods to run. I made up a quick example of such a rule on github, along with a test that uses it. My little example uses a system property for switching, but you can also obviously make it switch on anything you can think of that you can get your hands on here.
You can clone and run this example with:
git clone git#github.com:zzantozz/testbed tmp
cd tmp
mvn test -pl stackoverflow/9611070-toggleable-custom-ignore -q
mvn test -pl stackoverflow/9611070-toggleable-custom-ignore -q -D junit.invertIgnores
The one downside of this approach that I can think of is that your tests won't get properly marked as "ignored" because that's also done by the BlockJUnit4ClassRunner.runChild() method, and if you peek at ParentRunner.runLeaf() (which runChild() delegates to), you'll see that the notifier, which is what you need to report ignored tests, isn't passed down far enough to be used by a Rule. Again, it's something you'd need a custom Runner for.
You could create a special ant target that removes the #Ignore annotation and add an #ignore annotation to any active #Test annotated method
the target would be something like this:
<project name="anyname" default="test" basedir=".">
..
..
<target name="enable-ignored-test" depends="copy-code-to-replace-ignored">
<fileset id="fsTestCase" dir="." includes="**/*Test.java">
</fileset>
<replaceregexp flags="gm">
<regexp pattern="^#Ignore"/>
<substitution expression=""/>
<fileset refid="${fsTestCase}"/>
</replaceregexp>
<replaceregexp flags="gm">
<regexp pattern="#Test"/>
<substitution expression="#Ignore #Test"/>
<fileset refid="${fsTestCase}"/>
</replaceregexp>
</target>
<target name="run-ignored-tests" depends="enable-ignored-test,test" />
..
..
</project>
I'm trying to start a weblogic server using ant exec command,after the operation triggers the exec process creates a child process, i wish to kill the parent process and keep alive the child process.
My code below
<project name="startserver" default="start" basedir=".">
<target name="start">
<exec dir="C:\bea\user_projects\domains\devtest"
executable="cmd"
failonerror="true">
<arg value="/c"/>
<arg value="startWebLogic.cmd"/>
</exec>
</target>
</project>
this code starts server and my application is also up, problem is that ant script is still running & moreover it doesn't exit as shown below
how can i exit from the exec and kill the parent process, as a workaround i tried using
timeout
attribute of exec task, but it kills both parent and child process, how can i achieve this so that ANT exits and also server keeps running in the background.
Try the spawn="true" attribute of the exec task. It should do what you are looking for. But beware of the implications for input/output handling (see link).
When you set spawn="true" for exec tag, the child process runs in the the background even if current process ends execution. In your case instead of cmd, use startweblogic.cmd directly as exec command.